LECTURE 3

RITZ VARIATIONAL METHOD

(Weak Formulation)
Steps:
i) Bring all the terms of the governing
equation to one side of the equality
ii) Multiply with a weighting function
w(x)
iii)Integrate by parts over the limits of
the domain
iv)Separate linear and bilinear terms
v) Identify the boundary terms

2
 RITZ VARIATIONAL METHOD
(Weak Formulation)
Starting with the equation
d  du 
dx   (x) dx  - f(x) = 0 in 

The Weighted residue becomes

Xb
 d  du 
 w( x)   (x)  - f(x) dx = 0
Xa  dx  dx 
w(x) -- weighting function

i.e.,  R ( x ) w(x) dx
3
Observations:

u is differentiated twice, while w(x) is

remaining undifferentiated.
 So trial functions should be differentiable at
least twice.
But continuity of derivatives of higher order
is very difficult.
 Hence it is preferable to reduce the order
of derivatives of u as much as possible

4
We note that the first term is of the form

udv = uv  vdu

Where u = w(x)
 du 
And v =   (x) dx 
 
Xb Xb
 d du   du 

Xa
w( x) 
 dx
( (x) ) dx =  w(x) ( (x) )
dx   dx  Xa
Xb
du dw
-  ( x ) dx
Xa
dx dx
5
The equation can be now recast as
Xb Xb Xb
du dw   du  

Xa
 ( x)
dx dx
dx = 
Xa
f ( x) w(x) dx +  w(x)   (x)  
  dx   Xa

Xb
du dw
Now   (x) dx
Xa
dx dx
is a linear function of both field variable
and weighting function = B(u,w)

6
 Xb

And -  f ( x) w(x) dx
Xa

is a function of weighting function alone

  du  
Xb Represents the
 w(x)   (x) dx   boundary term where
    Xa
 du  Is the flux or secondary
  (x) dx 
variable

7
i.e., B(u,w) = ℓ(w)

B is the bilinear function and ℓ is the linear

function

Xb Xb Xb
du dw   du  

Xa
 ( x)
dx dx
dx = -  f ( x) w(x) dx +  w(x)   (x)  
Xa   dx   Xa

The above represents the weak form of the

original Governing equation
8
Recasting of the given differential equation
in this form where the order of derivatives
are traded between the trial function and
the weighting function, thereby weakening
the continuity requirement on the trial
functions is called ‘Weak Formulation’.
The original equation is recast into its
Weak Form.

9
In the Ritz method we take, w( x ) = u (x)
which implies that where ever u(x) is specified,
as at the boundary, w(x) = 0.

w( x) = u (x) Represents the variation of

the field variable.

10
APPLICATION OF VARIATIONAL
FORMULATION
Illustrative Example for Variational
Formulation

Consider the elastic deformation of

a tapered - rod under its weight and also due
to applied pull at the free-end, considered
previously.

11
The governing equation is
d du
[EA(x) ] + A(x) = 0 in 0 < x < L
dx dx

With B.Cs i) u(0) = 0

and
du
ii)At x=l [EA(x) ] P
dx

12
The WR formulation is
L d du
 w(x) { [ EA(x) ] +  A(x) } dx = 0
0 dx dx
where w(x) is the weighting function and
u(x) is the trial solution. Integrating by
parts and r-arranging, we get

L du dw L
 EA ( x ) dx =   A ( x ) w(x) dx - w(0) P(0) + w(L) P(L)
0 dx dx 0

i.e. B(u, w) = l (w)

13
since u(0) = 0 (specified), w = u at
x = 0 vanishes
i.e. w(0) = 0 P(L) = P - specified
Hence P(0) w(0) term vanishes

L du dw
B( u, w) =  EA ( x ) dx
0 dx dx
L
( w ) =   A(x) w(x) dx + Pw(L)
0

14
Since the bilinear term B is symmetric ie.
[B(u,w) = B (w,u)] a quadratic functional I(u)
exists and is given by I(u) = 1/2 B (u, u) - l(u)
L 1 du 2 L
I( u ) =  EA(x) dx -  A ( x ) u(x) dx -   u(L)
0 2 dx 0

strain-energy of deformation External work External workdone

by distributed load by concentrated
load
clearly I(u) gives the Total Potential of the
elastic system, which is stationary
L L
du du
 I(u) = 0 =  EA( x)  dx -   A(c) u(x) dx -   u(L)
0
dx dx 0
15
we know that w(x) = u(x) and threfore
 du  d dw
  = ( u) =
 dx  dx dx
 We get
L du dw L
 EA ( x ) dx =   A(x) w(x) dx - Pw(L)
0 dx dx 0

B( u, w ) = (w) - the weak form

16
 Advantages of the weak form
Order of the differential equation becomes
half of that in the original equation.

Hence continuity requirements on the

assumed solution is reduced.

Lower order polynomial can be assumed

for the approximate solution.
17
The Natural Boundary condition
becomes embedded in the weak form

Hence the trial solution needs to satisfy

only the essential boundary condition

18
 Ritz Method of Solution
2 3
u( x ) = a 0 + a1 x + a 2 x + a 3 x

Essential boundary condition is u(0) = 0

We get ao = 0 and
3
u ( x) =  a j  j (x) where  j (x) = x j

j=1

19
The weighting function is w(x) =i ( x) i = 1, 2, 3

substituting in the Weak-form of the governing

equation.
This leads us to the equation
3
di d j

j 1
a j  EA (x)
dx dx
dx = rj i = 1, 2, 3

L
where rj =   A(x)  j (x) dx + P  j (L)
0

20
on evaluation of the integral within the
brackets, this reduces to the set of
algebraic equations.
 k11 k12 k13  a1  r1 
k    
 21 k 22 k 23  a 2  = r2 
 k 31 k 32 k 33  a  r 
 3  3

L d i d  j
Where kij =  EA ( x ) dx
0 dx dx

21
Solution of this matrix equation leads to
determination of the constants a1,a2 and
a3 there by giving the approximate
solution.
3
j
u( x ) =  aj x
j=1

22
For the given illustrative example of a
tapered rod under its weight and also due to
applied pull at the free-end

when i = 1, j = 1. . . . k11
when i = 1, j = 2 . . . . k12
i = 1, j = 2 . . . . k13
and so on

23
 300
d1 d1
k11 =  EA(x) dx
0
dx dx
= E(80 - 0.2x).1.1dx = 1.5 x104 E
300
d1 d2
k12 =  EA(x) dx
0
dx dx

= E(80 - 0.2x).1.2x.dx = 3.6x106 E

300
d1 d3
k13 = 0
EA(x)
dx dx
dx

=E(80 - 0.2x).1.3x2 dx =8.6 x108 E

24
Similarly

k21 = 3.6x106 E k22 =1.2 x109 E

k23 = 2.88 x1011 E k31 = 8.6 x108 E

k32 = 2.88 x1011E k33 = 1.322 x1014 E

25
  A(x)  dx =  (80 - 0.2x) .x. dx = 1.3773 x 10
5
r1 = 1

r2 =   A(x)  dx =   (80 - 0.2x) .x . dx = 24 x 10

2
2 7

     3 9
r3 = A(x) 3 dx = (80 - 0.2x) .x . dx = 4.598 x 10
p1 = P. 1 (L) = P L = 3 x 10 7

p2 = P. 2 (L) = P L2 = 9 x 109
p1 = P. 3 (L) = P. L3 = 27 x 1011

1.5 x 10 4 3.6 x 106 9.45 x 108   a 1  1.37 x 105  3 x 107 

 6 9 11     7 9 
 3.6 x 10 1.2 x 10 2.88 x 10  *  a2  =  2.4 x 10 + 9 x 10 
 9.45 x 108 2.88 x 1011 1.322 x 1014   a3   4.598 x 109  27 x 1011 
  

26
On solving
-5
a1 = 6.6762 x 10
a2 = - 4.946 x 10 -8
a3 = 6.4736 x 10 -10

2 3
U | x 300 = a 1 (300) + a 2 (300) + a 3 (300) = 0.033056 cm

But the Strength of Material Method gives

deflection at the tip as = 0.0378 cm

27
THE FINITE ELEMENT METHOD
or NODAL APPROXIMATION METHOD:

The basic concept behind the Finite

element method is “going from part to
whole”
Name “FINITE ELEMENT” coined by
Clough
Fitting of a number of piecewise
continuous polynomials to approximate the
variation of the field variable over the entire
domain
28
STEPS INVOLVED IN THE FINITE ELEMENT
METHOD:

Discretisation of the structure: In this step the

given structure is divided into subdivisions or
elements. Depending upon the problem we
may choose I D, II D or IIID elements.

29
I D elements

30
Constant strain triangular element Bilinear Rectangular element

Linear strain triangular element Eight noded quadratic quadrilateral elemen

II D elements 31
 3
4

1 2

Linear Quadrilateral element

32
III D elements

33
Selection of suitable displacement
model:
We make an assumption as to the
variation of the unknown solutions over
the element. In general, the field variable
(example, temperature, displacement etc)
is assumed to vary linearly or quadratically
or cubically.

34
Displacement model associated with each
element

35
Field
variable
u

Length l 36
Derivation of elemental matrices and
load vectors:
From the assumed displacement model,
the elemental stiffness matrix [K]e and
load vector [P]e of the element are to be
derived using either equilibrium methods
or a suitable variational principle.

37
Assembly of elemental equations to obtain
overall stiffness matrix: the individual element
stiffness matrices and load vectors are to be
assembled in a suitable manner to get the
overall stiffness equation which is expressed
as
[K]{u} = {P}
where [K] is the assembled stiffness matrix
{u} is the vector of unknowns or nodal
displacements
{P} is the vector of nodal forces for the
complete structure
38
Imposition of boundary conditions: The
Boundary conditions could now be
incorporated to get the reduced equations.

Solutions for the unknown nodal

displacements: The elemental matrices,
on assembly, yield a set of equations,
which could be expressed as a set of
matrices, which could be solved using any
iterative procedure or numerical method.

39
Computation of elemental strains
and stresses: From the unknown
displacements, the element strains
and stresses can be computed by
using the necessary equations of
solid or structural mechanics.

40
 1

L1 = 10 cm 1

L2 = 10cm 2

E= 2x107N/cm2 2
EBC:

3
U1 = 0
A 1 = 2sq.cm
Pl = 10kN
A2=
1sq.cm 10 kN

41
[K]1 = EA1 1 –1
ℓ1 –1 1

[K]1 = 4 x 105 – 4 x 10 5
–4 x 105 4 x 105

[K]2 = EA2 1 –1
ℓ2 –1 1

[K]2 = 2 x 105 –2 x 105

–2 x 105 2 x 105
42
The assembled stiffness matrix is given by

4 –4 0
[K]g = 105 –4 4+2 -2
0 –2 2

43
 The load vectors are
{P}1 = R
0
where R is the reaction at the fixed end

{P}2 = 0
1

{P} = R
0
10
44
The overall equilibrium equation is given by
[K] {u} = {P}
or

2 x 105 2 -2 0 u1 R
-2 3 -1 u2 = 0
0 -1 1 u3 10

45
2 x 105 2 -2 0 u1 R
-2 3 -1 u2 = 0
0 -1 1 u3 10

2 x 105 3 -1 u2 0
-1 1 u3 = 1

u2 = 0.25 x 10-4 cm
u3 = 0.75 x 10-4 cm
46
Strain for element 1 = Є1
= u/ x for element 1
= (u2 – u1)/ ℓ1
= 0.25 x 10-5

Strain for element 2 = Є 2

= u/ x for element 2
= u3 – u2/ ℓ2
= 0.50 x 10-5

47
The stresses in the elements are given by

Stress in element 1 = 1 = Є 1E1

= (0.25 x 10-5) (2 x 107)
= 5 kN/cm2
Stress in element 2 = 2 = Є 2E2
= (0.50 x 10-5) (2 x 107)
= 10 kN/cm2

48
COMPUTATION OF REACTION AT FIXED END:

2 x 105 [2 * u1 – 2 * u2] = R
Substituting for u1and u2 we get
Reaction R= 10kN

49
NODAL APPROXIMATIONS
In general problems arise in
engineering where we seek an approximation
u (x, y, z) to some exact function u(x, y, z) to
any desired level of accuracy, i.e.

u (x, y, z) = u (x, y, z)

Many times the approximate function is

obtained as a series expansion of some
known function with undetermined
coefficients. e.g.
50
 n

u (x) =  i 0
ai xi (power series)
n

or u(x) = (a i cos ix + b i sin x) (Trigonometric series)

i 1

In these expansions ai - s are called the

“generalised coordinates”

51
u(xi) = ui i = 1,2,.....r. Forcing the
approximations to take on these specified
values at the specified points, we have
a1 
a 
 2
a3 
2  
ui = < 1 x i x i . . . x i n 1 > . 
. 
 
. 
a 
 n
(n x 1)
i = 1, 2, . . . . . r.
52
 u1  a 1 
Taking r = n. We have u  a 
 2  2
.  . 
{f i }   = [ Pn ]  
.  . 
.  . 
   
un  an 
(n x 1) (n x 1)

where the vector of ai s and matrix [Pn ] are

known

53
 a1  u 1 
There, if [Pn ] is non-singular,   u 
a
 2  2
.  1
. 
  = [Pn ]  
.  . 
.  . 
   
an  un 
and f ( x) =  1 x x 2 . . . . x n -1 > [Pn ] 1 {u}
(1 x n) (n x n) (n x 1)
u 1 
u 
 2

.  
=  N1 N 2 . . . . N n >  
. 
. 
 

u n 

54
(1 x n) (n x 1)
The last equation expresses the
approximation in terms of the function values
at selected points, as compared to the
expansion in terms of the generalised
coordinates.
 These selected points are called the
“nodal points” and {f} is called nodal-
variable vector.
 The functions Ni(x) are called the shape
functions.
 Finally u(x)=Ni(x)ui is called the Nodal
Approximation. Ni – s are also called as
55
Derivation of Shape function for two
=

noded element:
1) Let u(x) = ai + a2x in 0 < x < l
= <1 x> a1
a2
u(0) = u1 and u(1) = u 2

Therefore a1 = u1
a1 + a 2 l = u2

In matrix form 1 0 a1 = u1
1 l a2 u2
56
u1   a1 
=  1 0  
 
u 2  1
  * a 2 
a1  1 0
1
u1 
  =  *  
a 2  1  u 2 

= 1 0
 1 u1 
1 
 * u 
    2

57
u(x) = < 1 x >  1 0 u1 
 - 1/  
 1/ u2 

u1 
=  (1 - x/) x/ >  
u 2 
u1
= < N1 N2> = N1 u1 + N2u2
u2
N1 (x) = 1 - x/ N1 (0) = 1. N1 () = 0
N1 (x) = x/ N 2 (0) = 0. N1 () = 1
N1 + N2 = 1 58
It can be verified that

N i (x j ) = 0 i  j
= 1 i = j
=  ij
(Kronecker Delta Function)
N1 u2
N2
. u1

1 2 1 2

59
To provide for the possibility of a constant
or uniform field when f is constant at all
points in the domain
We have
n n
 f(x) = C = j=1
N i (x) f i = C  N i (x)
j=1

f1 = f2 = ……. = fn = c
n
 
j=1
N i (x) = 1

The above properties are very

important properties of shape
functions.
60
1
1

1 X=0 2 x=l

X
=
l

61
 2
2) a
Let u(x) = 1 + a 2 x + a 3 x
Shape functions for a1 
 
quadratic elements
=  1 x x2 > a 2 
a 
 3
Taking x1 = 0, x2 = l/2 , x3 = l We have
u1  1 0 0 a1 
     
u2 
2
1 /2  /4  a 2 
u 
= 1 a 
  2   3
 3

a1   1 0 0 u1 
     
a 2  =  - 3/ 4/  1/   u2 
a   2 /2 u 
 3  - 4/2 2/2   3
62
 u1 
u(x) = < N1 N2 N3 >  
u2 
u 
 3

N1 (x) = 1 - 3x/ + 2x 2 /2

N2 (x) = 4x/ - 4x 2 /2
2 2
N3 (x) = - 3/ + 2x /

N1 (0) = 1 N1 (/2) = 0 N1 () = 0

N1 (0) = 1 N1 (/2) = 0 N1 () = 0
N1 (0) = 1 N1 (/2) = 0 N1 () = 0
N1 + N2 + N3 = 1 63
64
 Finite Element Formulation
• In FEA, we use the nodal approximation to
specify the unknown function in terms of its
values at selected ‘nodal points’, through a
Nodal Approximation
n
u ( x)  N j ( x)u j where
j 1

N j  s are the " Interpolating " or " shape " functions

u j  s are the values of ' u ' t these nodal po int s
It is seen that the shape functions automatically
satisfy the specified essential boundary conditions
The weighting functions are chosen from the shape
functions;  ( x)  N i ( x) i 1, 2,...n 65
The governing equation is
d du
[EA(x) ] + A(x) = 0 in 0 < x < L
dx dx

With B.Cs i) u(0) = 0

and
du
ii)At x=l [EA(x) ] P
dx

66
 Weak form is given by
L L
du dw

0
EA( x)
dx dx
dx =  A(x) w(x) dx  P(L)w(L) - P(0) w(0)
0

Substituting in the weak form

u(x) = N1u1 +N2u2

And w(x) as N1 first and then N2 we get a

system of two equations in two unknowns
namely u1 and u2

67
 e e e
[K ] [u ] = | r |
he
dN i dN j
K e
ij = EA(x) ( x) dx
0
dx dx

he

rje = A( x) N
0
j dx + Pj

68
 l
dN1 dN1
K e
11 = EA(x) dx
0
dx dx

= ∫ E A (-1/l)(-1/l)dx
= EA/l2 ∫ dx
= EA/l

69
 l
dN 1 dN 2
K e
12 = EA(x) dx
0
dx dx
= ∫ E A (-1/l)(1/l)dx
= -EA/l2 ∫ dx = - EA/l
r 1  l 2 A1  A2 
   
 r2  6  2 A2  A 1 

= ∫ E A (-1/l)(1/l)dx
= -EA/l2 ∫ dx = - EA/l
70
 l
dN 2 dN 2
K e
22 = EA(x) dx
0
dx dx

= ∫ E A (1/l)(1/l)dx
= EA/l2 ∫ dx
= EA/l
Stiffness matrix for 2 noded element
K = EA 1 -1
l -1 1
71
 he

r je = AN
0
j dx + Pj

he

r1e = AN
0
1 dx = γ Al/2

he

r2e = AN
0
2 dx = γ Al/2

{r}= γ Al/2 1
1 72
A(x) = A1 – (A1- A2) x/l
ie.A(x) = 80 – (80-20)x/300
= (80 – 0.2x)

 = 0.075 N/cm3
E = 2 x 107 N/cm2

73
If for the entire domain, there are only two
nodal points, they also happen to be the
boundary points x = 0 and x = L n = 2 and ij =
1, 2. The above equation reduces to

u1  r1 
  =  
[K] u2  r2 

2x2 2x1 2x1

74
 Example
Consider the tapered rod problem
 = 0.075 N/cm3 L =300cm
E = 2 x 107 N/cm2
u1 = 0 P1 = R
P2 = P

N1(x) = 1 – x / L N2(x) = x / L

dN 1 1 dN 2 1
= - = -
dx L dx L
75
A(x) = 80 – 0.2x
2
300
 1 E
K11 =  E (80 - 0.2x)  -  dx = L2 (80-.02x) dx
0  L
E 0.2L2
= 2
[80L - ]
L 2
E 50E E
= = =
(80 - 0.1 L) 300 6
L
E
K12 = K 21 = 
6
= E
K 22 =
6 76
 E 1 -1 
[K] =  -1
6  1
L
r1 =  (80 - 0.2x) (1- x / L) dx + R
0

= 675 + R

300

r2 =   (80 - 0.2x) ( x/L) dx + P

0

= 450 + 10 5
77
Apply the Boundary Condition u1 = 0, this
reduces to

k22u2 = r2 r
2
u2= K 22 = 0.03 cm

This is the value of a uniform rod

with average area under the pull. This
compares with the Ritz method discussed
earlier with a cubic polynomial which worked
out to

u2= 0.033056 cm 78
ao  a1 x u ( x)
Linear displacement model
ao  a1 x  a2 x 2 u ( x)
quadratic displacement model
ao  a1 x  a2 x 2  a3 x 3 u ( x)

cubic displacement model

79